A Quasi-Newton Neural Network Based Efficient Intrusion Detection System for Wireless Sensor Network

Abstract

In Wireless Sensor Networks (WSN), attacks mostly aim in limiting or eliminating the capability of the network to do its normal function. Detecting this misbehaviour is a demanding issue. And so far the prevailing research methods show poor performance. AQN3 centred efficient Intrusion Detection Systems (IDS) is proposed in WSN to ameliorate the performance. The proposed system encompasses Data Gathering (DG) in WSN as well as Intrusion Detection (ID) phases. In DG, the Sensor Nodes (SN) is formed as clusters in the WSN and the Distance-based Fruit Fly Fuzzy c-means (DFFF) algorithm chooses the Cluster Head (CH). Then, the data is amassed by the discovered path. Next, it is tested with the trained IDS. The IDS encompasses ‘3’ steps: pre-processing, matrix reduction, and classification. In pre-processing, the data is organized in a clear format. Then, attributes are presented on the matrix format and the ELDA (entropy-based linear discriminant analysis) lessens the matrix values. Next, the output as of the matrix reduction is inputted to the QN3 classifier, which classifies the denial-of-services (DoS), Remotes to Local (R2L), Users to Root (U2R), and probes into attacked or Normal data. In an experimental estimation, the proposed algorithm’s performance is contrasted with the prevailing algorithms. The proposed work attains an enhanced outcome than the prevailing methods.